General:

I just remembered the trailer for the game and thought it would be interesting to see how a lora would work based on it.

Since the game is not even out yet and there was just the one trailer to work on it turned out as you may expect - pretty bad. For some reason even though the trailer itself was available with quality up to like 4k, despite also cropping it with minimal downscaling the dataset wasn't very sharp, so the generations come out blurry too.

For this version I tried training two characters (player and the character named Sayuki) and the style itself - probably a bad idea generally.

It's just an experiment - it's pretty bad too, so I wouldn't expect this to go anywhere, unless the game comes out.

Tags and information

Use tags:

@varsapura_style, 3d

For characters:

player:
vsp_plrchar, 1girl, black hair, wolf cut, green eyes, grey jacket, open jacket, grey collared shirt, black shirt, black belt, skirt

sayuki:

vsp_s4yki, 1girl, brown hair, brown eyes, police uniform, suspenders, beret, white shirt, necktie, armband, sleeves rolled up, blue skirt, thigh strap, high heels

Some info:

Aside from the blurriness and lack of texture issue (maybe you could use other 3d loras to add some definition - I tried my Madruga3d lora, and it did help a bit - 0.3 weight and without using the trigger words - see example images).

Other issue is the policewoman and police uniform concept kind of leans towards the look of the Sayuki character (with the suspenders, badge, id card, and whatnot - you may need to put some features to negative if you get what you haven't prompted for). Lastly, the outfit details aren't exact or consistent, so keep that in mind, and for whatever reason, high heels can turn to high heel boots (you can mitigate by using "pumps" alongside it)

Support:

If you like what you find here, please consider making a donation https://ko-fi.com/hat0ri, thanks!